Study on RF Characteristics and

Modeling of Scaled MOSFET

Presented by:

05M36370 Masayuki Nakagawa

Supervisor:

Professor Hiroshi Iwai

Department of Electronics and Applied Physics

Interdisciplinary of Graduate School of Science and Engineering
Tokyo Institute of Technology
2007, January
Contents
Chapter 1 Introduction ---------------------------------------------------- 1
1.1 RF CMOS Technology ---------------------------------------------------------- 1
1.1.1 RF CMOS Applications ------------------------------------------------ 2
1.2 The Indicators of RF Device --------------------------------------------------- 3
1.2.1 The Cut-off Frequency ------------------------------------------------- 4
1.2.2 The Maximum Frequency --------------------------------------------- 4
1.2.3 The Noise Figure -------------------------------------------------------- 5
1.3 Requirements of RF Characteristics ---------------------------------------- 6
1.4 High-k Gate Materials ---------------------------------------------------------- 6
1.4.1 Dielectric Dissipation --------------------------------------------------- 8
1.5 Purpose of This Study --------------------------------------------------------- 10

Chapter 2 Methodology of RF Measurement ---------------- 11

2.1 RF Measurement --------------------------------------------------------------- 11
2.1.1 de-embedding ----------------------------------------------------------- 12
2.2 Noise Figure Measurement -------------------------------------------------- 13
2.3 Modeling of DC Model -------------------------------------------------------- 16
2.3.1 Threshold Voltage Model --------------------------------------------- 16
2.3.2 Essential Equation for I-V Characteristics ----------------------- 17
2.3.3 Channel Charge Density Model ------------------------------------ 19
2.3.4 Mobility Model --------------------------------------------------------- 21
2.3.5 I-V Model in the Strong Inversion Region------------------------ 23
2.3.6 Subthreshold I-V Model ---------------------------------------------- 26

i
 2.3.7 Other Important Parameters --------------------------------------- 28
2.3.8 Source/Drain Series Resistance ------------------------------------- 29
2.4 Modeling of the Extrinsic MOSFET --------------------------------------- 34
2.4.1 Gate Resistance Model ----------------------------------------------- 35

Chapter 3 RF characteristics of High-k MOSFET --------39

3.1 device structure ---------------------------------------------------------------- 39
3.2 DC characteristics ------------------------------------------------------------- 42
3.3 RF characteristics ------------------------------------------------------------- 45
3.3.1 Cut-off Frequency and Maximum Oscillation Frequency ----45
3.4 Parameter Extraction -------------------------------------------------------- 52
3.4.1 Extraction of DC parameter ---------------------------------------- 52
3.4.2 Extraction of Extrinsic Components by Using Equivalent
Circuit ------------------------------------------------------------------- 54
3.4.3 Influence of Gate Leakage Current to Gate Resistance ------- 56
3.4.4 fT and fmax in conjunction with Rg ------------------------------- 61
3.4.5 Analysis of Gate capacitance ---------------------------------------- 62

Chapter 4 Exploration for Optimum Device Structutre --- 66

4.1 Device Structure ---------------------------------------------------------------- 66
4.2 DC Characteristics ------------------------------------------------------------ 67
4.3 RF Characteristics ------------------------------------------------------------- 73
4.3.1 Finger Length Dependency of fT and fmax ---------------- 73
4.3.2 Noise Characteristics ---------------------------------------- 74

Chapter 5 Conclusions and Future Issue ---------------------- 76

ii
Reference ------------------------------------------------------------------------ 78

Acknowledgements -------------------------------------------------------- 81

iii
Chapter 1
Introduction
1.1 RF CMOS Technology

With recent fast growth in the RF(Radio-Frequency) wireless communications

market, the demand for high performance but low cost RF solutions is rising. And with

the rapid growth of CMOS (Complementary Metal Oxide Semiconductor) technology

due to its scaling, it comes to be considered as the mainstream of RF technology due to

its low cost, low power and high integration. This advanced performance of CMOS is

attractive for RF circuit design in view of a system-on-chip realization, where digital,

mixed-signal baseband, and RF transceiver blocks would be integrated on a single chip.

RF CMOS application is discussed in section 1.1.1.

However, with extreme scaling down of CMOS, several serious issues, e.g.

short-channel and narrow width effects, impact ionization and gate leakage current, etc.

have come out. Especially, gate leakage current is the most serious issue, because it

makes difficult to realize Low Stand-by Power (LSTP) of CMOS devices. To improve

this issue, high dielectric constant materials, so-called High-k, for gate oxide as

replacement for conventional SiO2 have attracted attention. High-k technology is

discussed in section 1.1.2.

1
1.1.1 RF CMOS Applications

Figure 1.1 Application spectrum, ITRS2005

Recently, with the development of a highly-information-oriented society, the RF

(Radio-Frequency) wireless communications market dramatically grows up. In the

future, the demands of RF wireless technology will continue to increase all the more in

various areas, e.g. our daily life, industrials and medicals. These demands are realized

by the development of semiconductor products. Beforetime, the compound

semiconductors composed of elements from group and in the periodic table such

as SiGe, GaAs and InP had been mainly dominant for high-speed communications

and RF applications as shown in Figure 1 [1]. However, RF CMOS application has

been researched and developed by Universities in Europe and the United States cheifly

[2][3], and then RF CMOS technology gets to be considered as the mainstream due to

its low cost, low power, high integration and easy access to the technology, even

though the compound semiconductors still dominate out of 5GHz as shown in Figure

1 [4]. Since RF characteristics, such as the cut-off frequency (fT) and the maximum

oscillation frequency (fmax), of Si ULSI come to advance up to several dozen GHz (see

2
Tabel 1), available bandwidth increases and the signal frequency transmitted to

interconnection of LSI comes to over GHz. Thus, CMOS technology comes to be

important for both the RF technology of wireless applications and digital LSI

applications. The feature that CMOS circuit configuration and multi-layer

interconnections are available for RF applications are also attractive. However, there

are some challenges for RF CMOS applications in the following:

1. The necessity of optimization at semi-restricted process conditions.

2. The difficulty of adopting high resistive substrate.

The solution of these challenges is necessary for RF CMOS applications.

Table 1.1 The roadmap of RF CMOS characteristics [5]

Year 1995 1997 1999 2001 2003 2005 2007 2009
Gate length [nm] 250 180 140 120 100 70 50 35
Gate width [um] 200 150 110 100 80 80 50 35
fT[GHz] 39 50 65 80 105 145 205 420
fmax[GHz] 39 42 46 50 60 62 68 85
NFmin@2GHz 0.3 0.26 0.22 0.17 0.14 0.13 0.1 0.08

1.2 The Indicators of RF Device

In evaluating device RF performance, some indicators are employed as the cut-off

frequency fT, the maximum oscillation frequency fmax, the minimum Noise Figure NFmin

(and sometimes 1/f noise). In this section this three indicators are discussed especially.

3
1.2.1 The Cut-off Frequency fT

fT is defined as the transition frequency which small-signal current gain drops to unity.
It is a measure of the maximum useful frequency of a transistor when it is used as an

amplifier. fT can be easily obtained by converting measured S21 parameter to H21

parameter.
gm
fT = (1.1)
2πC gs

1.2.2 The Maximum Oscillation Frequency fmax

fT is surely a good indicator of the low-current forward transit time. However, as a

performance indicator, it does not include the effects of gate resistance Rg, which are

very important in determining the transient response of a transistor. So a indicator

including Rg effects have been proposed, fmax. fmax is the frequency at which the

unilateral power gain becoms unity.

2
1 S 21
−1
2 S12
UnilateralGain = ; K=Kfactor (1.2)
S ⎛S ⎞
K 21 − Re⎜⎜ 21 ⎟⎟
S12 ⎝ S12 ⎠
fT
f max = (1.3)
2 g ds ( Rg + Rs ) + 2πfT RgCgd

4
1.2.3 The Noise Figure NFmin

The RF noise Figure is an important parameter. Due to the low effective gate

resistance Rgate and the high fT the minimum noise Figure is extremely small at

frequencies of few GHz, and the measurements have to be accurately de-embedded and

extended to get a proper picture.

Noise can be defined as any interference unrelated to the signal of interest. The most

commonly accepted definition for NF is given by Eq. (1.4), where SNRin and SNRout are

the signal-to-noise ratios measured at the input and output, respectively.

SNRin
NF = (1.4)
SNRout T =300 K

NF is a measure of the degradation of the SNR as the signal passes through a system.

For a noiseless circuit SNRin = SNRout . Therefore, regardless of the gain, NF equals to

the unity. However in real systems the inner noise degrades the SNR, yielding NF>1.

NF for circuits working in voltage-mode becomes

Vn2,out
NF = (1.5)
Av2 N RS

For current-mode circuits NF becomes:

I n2,out
NF = (1.6)
Ai2 N RS

Where Vn2,out =voltage of the total output-noise, I n2,out = current of the total output-noise,

Av= voltage gain, Ai= current gain, and NRS=noise of the resistance of a source.

5
Since noise performance is an important design factor regardless of the technology used,

improving the noise Figure of a device is desirable.

1.3 Requirements of RF Characteristics

As a reference for the performance of the RF CMOS for circuit application, there are
specific requirements. For many RF applications, such as PA drivers or wireless LAN

PA’s require power level over 20 dBm. Table 1.2 shows the requirements for wireless

communication system.

Table 1.2 Requirement for wireless communication system

PAN/LAN Cellular Phone

Bluetooth 802.11b 802.11a/g PDC W-CDMA

Frequency(GHz) 2.4 2.4 5.2/2.4 0.8 2.1

Power Output(dBm) 0/20 20 20 29 24

1.4 High-k Gate Materials

As mentioned above, with its extreme scaling-down of CMOS devices, CMOS

technology come to play an important role in digital and/or analog applications.

However, MOS structure device has a rule, so-called the scaling rules, which make

physical effects on device complex and difficult. Table 1.3 shows the scaling rules for

various device parameters. In this table, scaling factor is k>1.

6
 Table 1.3 Miniaturization with scaling factor of k[6]
Parameter Initital Scaled
channel length L L/k
channel width W W/k
total device area A A/k2
gate oxide thickness Tox Tox/k
gate capacitance Cox Cox×k
junction depth Xj Xj/k
power supply voltage Vdd Vdd/k
threshold voltage Vth Vth/k
substrate doping concentration NSUB NSUB×k
S/D doping concentration NS/D NS/D×k

As to gate oxide thickness Tox which plays an important role in MOS structure device,

with scaling aggressively, Tox become too thin and the gate leakage current density

will become increase as shown in Figure 1.2. This is the most serious issue in Si MOS

device.

To control the gate leakage current, many materials as replacaement for SiO2 which

have a high dielectric constant, have been researched. Other keywords for High-k are

its band gap, band alignment to silicon, thermodynamic stability, film morphology,

interface quality, process compatibility and reliability. In this thesis, sub-100 nm RF

CMOS with HfSiON high-k gate dielectrics will be evaluated.

7
Figure 1.2 Prediction of equivalent oxide thickness and the maximum gate leakage

current density variation[4]

1.4.1 Dielectric Dissipation

As important concern of high-k at RF region, dielectric dispersion is included. High-k

may not keep the high dielectric constant at RF region, that is to say, high-k is not

“high-k” at the region. Then, following mentions dielectric dispersion quoting from [7].

Frequency dependency of dielectric function is called dielectric dispersion generally. As

for the dispersion where electronic polarization and ion polarization relate resonance

type, or, oscillator dispersion is shown, but as for the dispersion where orientation

polarization relates diapersion of relaxation type is shown as relaxation function of

Debye type shows. Then, the dispersion of the microwave range which orientation

polarization relates is especially called dielectric relaxation. The real part and imaginary

part of dielectric function which respective polarization relates are shown in Figure 1.3

[8].

8
 Electron Polarization
Orientation Polarization
Orientation Polarization
Ionic Polarization
Ionic Polarization

9
Electron Polarization
Micro wave region Infrared Region Ultraviolet Region
Figure 1.3 Dielectric dissipation [7],[8]
1.5 Purpose of This Study

In this thesis, device structure of next generation MOSFET including gate insulator and
gate finger layout for maximizing the performance in RF circuit application is explored.

Measuring S-parameter up to 40 GHz and Noise Fiugre of scaled MOSFET, we

analyzed cut-off frequency fT, maximum oscillation frequency fmax and minimum Noise

Figure NFmin as indicators of RF performance based on extracted parameters by DC

modeling and equivalent circuit.

In terms of solutions on the material of gate insulator, high-k material HfSiON is

studied. High-k is now well researched and developed. However, report on RF

application of high-k device is little until now. Although discussion about applying

high-k to RF MOSFET is one of the purposes in this study, high-k MOSFET in RF

region has some concerns. Those concerns are discussed about in Chapter 3.

Then, in terms of solutions on the structure which means the structure of gate finger on

multi finger device for RF circuit, devices which have various finger length is studied.

From the standpoints described above which are material of gate insulator and structure

of gate finger, we explore the device solutions for RF circuit.

10
Chapter 2
Methodology of RF Measurement
2.1 S-parameter Measurement

Recently, with the rapid growth of RF (Radio-Frequency) wireless communications

market, the requirements for RF applications increase. Therefore, at that stage of

development, accurate RF measurement must be performed. It is well-known that

accurate RF measurement on wafer is difficult due to parasitic components of system

and device. Moreover, the effect of the surrounding parasitic components come to be

bigger for device characteristics with continuous downscaling of the device dimensions.

So two-step correction procedure has to be done as below:

1. Calibration

- The measurement system has to be calibrated, defining a reference plane for

the S-parameter measurements at the probe tips using a standard calibration

technique (SOLT, LRM etc.).

2. De-embedding

- The surrounding parasitic components on wafer have to be characterized, so

that intrinsic S-parameters of transistor can be obtained from the

measurement.

If calibration is performed exactly, S-parameters of inner reference plane can be

obtained. Secondly, device parasitic components must be de-embedded which

11
methodology is discussed in the next section.

2.1.1 De-embedding

Resistance, capacitance or inductance included in the interconnect lines or bonding

pads, which could neglect approximately in DC measurements (not completely neglect

as for the interconnect lines resistance), largely influence to device RF performance.

These parasitic components must be removed to evaluate intrinsic device characteristics

at high frequency. De-embedding is the method of removing parasitic components.

For sub-micron transistors, the extraction is very sensitive to the input/output pads

de-embedding. If the de-embedding is not complete, the correct values of the equivalent

circuit components from direct extraction can not be obtained. Different de-embedding

techniques have been developed based on different calibration test structures [9]-[12].

De-embedding procedure in this study is described as bellow. At first, Figure 2.1 shows

OPEN and SHORT TEG (Test Element Group) for removing parasitic components of

pad and interconnect lines.

Source Pad Source Pad

Gate Drain Gate Drain

Source Pad Source Pad

(a) OPEN (b) SHORT

Figure 2.1 OPEN and SHORT TEG

12
As mentioned above, parasitic components are included in crude measured data. Then,

parasitic capacitance of pads can be de-embedded by subtracting measured S-parameter

of OPEN from crude measured S-parameter of DUT. Moreover, contact resistance can

be de-embedded from measured S-parameter of SHORT. Finally, S-parameter of only

device not including parasitic components can be gotten.

2.2 Noise Figure Measurement

Noise Figure measurement set up is shown in Figure 2.2.

Figure 2.2 Noise Figure measurement set up

Noise measurements allow the determination of the four Noise Parameter of a device

(transistor). For this concept four areas should be explained:

13
z Noise Parameters

z Noise Measurements

z Effect of second stage on noise behavior

z Noise Temperature

Noise Parameters:

These four numbers fully describe the noise behavior of an active or passive device (two

port) at a given frequency. For practical reasons, following quantities are used as Noise

Parameters:

z Minimum Noise Figure (NFmin): This is the small Noise Figure that the device can

reach at a given frequency and bias, if it is optimally matched at the source.

z Equivalent Noise Resistance (Rn): This is a number in Ωdimension that indicates

how fast the Noise Figure increases when the input (source) is mismatched.

z Optimum Noise Reflection Factor (Γopt): This is often used also as Optimum

Admittance Yopt: Is the source admittance required for the DUT to perform NFmin

(Yopt = Gopt+jBopt), 2 parameters.

The Noise Figure does not depend on the Load impedance presented to the device. It

only depends on the source impedance.

There exists a simple relation between the four Noise Parameters:

Rn 2
Noise Figure NF (YS ) = NFmin + ⋅ YS − Yopt (2.1)
GS
where Ys=Gs+jBs. This is the equation of a set of Circles on the Smith Chart (Noise

14
Circles) for which the value of the Noise Figure is the Level on each Circle.

This Circle Representation is only possible because the transistors Noise behavior is a

Small Signal Phenomenon.

Measurement of the Noise Parameter:

The four Noise Parameters can be determined if the Noise Figure of a device are

measured at a minimum of four different source impedances.

Because of errors associated with the extremely low-level noise measurements (the

typical noise power of a transistor is about -110 dBm in 1 MHz bandwidth = the Noise

Power is directly proportional to the Bandwidth).More than 4 impedances are in general

measured and the measured data are averaged. 7 to10 impedances are typically

sufficient for the determination of the noise parameters at each frequency.

The setup required to measure the Noise Parameter includes: - A Noise Analyzer or a

Spectrum Analyzer- A Noise Source – A Programmable Tuner – A mixer and Local

Oscillator (for f> 1.8 GHz) – Isolations, bias tees, etc.

Contrary to Load Pull measurements, it is not absolutely required that the Tuner used in

Noise Measurements to be pre-calibrated. It can be set to a number of positions and

readings that can be taken, both of the Noise Figures and the Tuner positions. If the

tuner is then characterized at those positions using a Network Analyzer, the 4 Noise

Parameters can be calculated. This is possible only because of Eq. (2.1). Of course, if

the tuner is pre-calibrated, this facilitates a lot the operation and the result can be

computed immediately. Again, a non pre-calibrate tuner well not permit to tune to the

minimum Noise Figure; it will only permit to compute it.

15
2.3 Modeling of DC Model

Since BSIM3v3 (Berkley Short-channel IGFET Model 3 version 3) develped by UCB

(University of California, Berkley) had been chosen for a standard compact model in

1995, this model have been widely used for digital and analog circuit design. A

philosophy embedded in the BSIM3 model is to find a physically accurate functional

form to describes a physical phenomenon and then use fitting parameters and even

empirical terms to achieve quantitative match with the device characteristics. In this

chapter, several models for DC parameter extraction such as thresholod voltage model

and I-V models etc. in BSIM3v3 are described. In addition to this, extrinsic parameters

extraction methodologies for a sub-circuit model are also described. With recent fast

growth in the radio-frequency (RF) wireless communications market, the requirements

for an accurate RF compact model rapidly increase.

2.3.1 Threshold Voltage Model

Accurate modeling of the threshold voltage (Vth) is one of the most important

requirements for the precise description of a device electrical characteristics. By using

the threshold voltage, the device operation can be divided into three operational regions.

If the gate voltage is much larger than Vth, the MOSFETs is operating in the strong

inversion region and the drift current is dominant. If the gate voltage is much less than

Vth, the MOSFET operates in the weak inversion or subthreshold region and diffusion

current is dominant. If the gate voltage is very close to Vth, the MOSFET operates in the

transition region called moderate inversion where both diffusion and drift currents are

important.

The complete Vth expression in BSIM3v3 is as following [17]:

16
 (2.2)

In this equation, the second and third terms are used to model the verical non-uniform

doping effect, the fourth term is for the lateral non-uniform doping effect, the fifth term

is for the narrow width effect, the sixth and seventh terms are related to the short

channel effect due to DIBL, and the last term is to descibe the small size effect in

devices with both small channel length and small width. A simpler model for Vth would

be preferred if it could offer the same adequate accuracy. In this work, since the fifth,

sixth and last terms are not considered for DC parameter extraction, only considering

terms will be introduced.

2.3.2 Essential Equation for I-V Characteristics

A DC MOSFET model is derived based on

(2.3)

(2.4)

17
Jn and Jp are the current dinsities for electrons and holes respectively, q is the electron

charge, μn, μp are the mobility of electrons and holes respectively, n and p are the

electorn and hole concentrations, E is the electric field and Dn and Dp are the diffusion

coefficients of electrons and holes, respectively. Dn and Dp are linked to μn and μp with the

following Einstein’s relationship:

(2.5)

(2.6)

where vt is the thermal voltage.

The terms of E in Eqs.(2.5),(2.6) represent the drift current components due to the

electric field E. The second terms of Eqs.(2.5),(2.6) describe the diffusion current

components due to the carrier cancentration gradient. In the strong inversion region, the

current is dominated by the drift current. In the subthreshold region, the diffusion

current component dominates. However, in the transition region (moderate inversion

region) from subt-hreshold to s-trong inversion, both drift and diffusiion currents are

important. As shown in Eqs.(2.5),(2.6), carrier density and the velocity-field

relationship are two fundamental factors determining the I-V characteristics. We need to

model the channel charge and mobility as well as the velocity-field relationship

carefully to describe the current characteristics accurately and physically. We discuss

the modeling of channel charge and mobility next before we introduce the modeling the

I-V behavior

18
2.3.3 Channel Charge Density Model

For the weak inversion and the strong inversion regions, separate expressions for
channel charge density have been given respectively:

(2.7)

(2.8)

We can combine these expressions in the following form:

(2.9)

(2.10)

As shown in Figure 2.3 and 2.4, Vgsteff becomes Vgs-Vth in the strong inversion region,

qε Si N CH vt V − Vth − VOFF
and follows exp( gs ) in the subthreshold region.
4φ B Cox nvt

The form of Eq.(2.10) was chosen to obtain a continuous equation for the channel

charge and to match the measured Qchs-Vgs characteristics in the moderate inversion

(transition) region. The channel charge expression, Qchs0, will be used in subsequent

sections of this chapter to model the drain current. Furthermore, the model accurately

predicts the charge in the transition (moderate inversion) region. The continuous and

accurate nature of the model makes it very attractive and promising in circuit simulation

since the moderate inversion region is becoming more important for low voltage/power

circuit application.

19
Figure 2.3 Vgsteff vs. Vgs-Vth in a linear plt [18].

20
2.3.4 Mobility Model
A good model for the carrier surface mobility is critical to the accuracy of a MOSFET

model. The scattering mechanisims responsible for the surface mobility include

phonons, coulombic scattering, and surface roughness scattering. For good quality

interfaces, phonon scattering is the dominant scattering mechanism at room temperature.

In general, mobility depends on many process parameters and bias conditions. For

example, mobility depends on the gate oxide thickness, doping concentration, threshold

voltage, gate voltage and substate voltage, etc.

Figure 2.4 Vgsteff vs. Vgs-Vth in a semi-logarithmic plot [18].

The continuity of mobility model is also required to ensure the continuity of the I-V

model. To achieve continuity in the mobility model, BSIM3v3 uses a unified mobility

expression based on the Vgsteff expression of Eq.(2.10),

21
 (2.11)

where

(2.12)

It can be seen that Eq.(2.11) follows Eq.(2.12) in strong inversion, and becomes a

constant in the subthreshold region. Several mobility model options are provided for

users to choose in BSIM3v3. a selector parameter called mobMod is introduced for this

purpose. The mobility expression in Eq.(2.11) has been designated as mobMod=1. The

following empirical mobility model option (mobMod=2) is better suited for depletion

mode devices:

(2.13)

BSIM3v3 also introduced a third mobility model option (mobMod=3):

(2.14)

It is clear that all of the mobility models given above approach constant values that are

independent of Vg when Vgs¡Vth. It should be pointed out that all of the mobility models

given above account for only the influence of the vertical electrical field. The influence

of the lateral electrical field on the mobility will be considered when discussing the

velocity saturation effect in the next section.

22
2.3.5 I-V Model in the Strong Inversion Region

2.3.5.1 I-V Model in the linear (triode) Region

1. Intrinsic case (Rds=0)
In the strong inversion region, the current equation at any point y along the channel is

(2.15)

where Vgs =(Vgs-Vth), Weff is effective device channel width. Cox is the gate capacitance
t

per unit area. V(y) is the potential difference between the channel and the source. Abulk

is the codfficient accounting for the bulk charge effect and v(y) is the velocity of

carriers. BSIM3 I-V formulaition starts with a simple piece-wise saturation velocity

model,

(2.16)

(2.17)

where Ey is the magnitude of the lateral electric field and Esat is the critical electric field

at which the carrier velocity becomes satureted. μis the mobility including the influence

of the lateral electric field Ey and is given by

(2.18)

In order to have a continuous velocity model at Ey=Esat, Esat satisfies

23
 (2.19)

Thus, before the electric field reaches Esat the drain current can be expressed

(2.20)

Eq.(2.20) canbe written as

(2.21)

By integratin Eq.(2.21) from y=0 to y=Leff, the effective channel length, and V(y)=0 to

V(y)=Vds, we arrive at

(2.22)

The drain current model in Eq. (2.22) is valid before the carrier velocity saturates, that is,

in the linear of the triode region.

2. Extrinsic case (Rds>0)

The parasitic source/drain resistance is an important device parameter which can affect

MOSFET performance significantly in short channel devices. The most straightforward

and accurate way of modeling the parasitic resistance effect is to use a circuit with

24
resistors in series with the intrinsic MOSFET. This leads to a complicated drain current

expression. In order to make the model efficient, the drain current in the linear region

can be modeled by extending Eq.(2.22).

(2.23)

where Ids0 is the intrinsic current expression given by Eq.(2.22). Rds is a variable to

account for the influence of the parasitic resistances at the source and drain.

2.3.5.2 Drain Voltage at Current Saturation, Vsat

1. Intrinsic case (Rds=0)

If the drain voltage (and hence the lateral electric field) is sufficiently high, the carrier

velocity near the drain saturates. The channel may be divided into two portions: one

adjacent to the source where the carrier velocity is field-dependent and the others

adjacent to the drain where the velocity has satureted. At the boudary between the two

portions, the channel voltage is the saturation voltage (Vdsat) and the lateral delctric field

is equal to Esat. We can substitute v=vsat and Vds=Vdsat into Eq.(2.15) to obtain the

saturation current:

(2.24)

By equating Eq.(2.15) and (2.24) at Vds=Vdsat, we can solve for the saturation voltage

25
Vdsat:

(2.25)

2. Extrinsic case (Rds>0)

Due to the parasitic resistance, the saturation voltage Vdsat will be larger than what is

predicted

by Eq.(2.25). Equating Eq.(2.23) with Eq.(2.24), Vdsat with parasitic resistance Rds will

be

(2.26)

(2.27)

(2.28)
(2.29)

λ=A1+Vgst+A2 is introduced to account for non-saturating effect of the device I-V which

will be discussed in 2.2.6.

2.3.6 Subthreshold I-V Model

In the subthreshold region, the potential in the channel exhibits a peak between the

source and the drain. At or near the peak of the potential, the lateral electric field can be

considered zero because the potential gradient is zero. Thus, the drift current can be

ignored in the subthreshold region. According to the current density equation given in

Eqs.(2.3) and (2.4), we have:

26
 (2.30)

We would like to use the charge sheet dinsity expression Qinv in modeling the I-V

characteristics. If we integrate Eq.(2.30) from the Si-SiO2 interface to the edge of the

depletion layer (Xdep) in the bulk, the current in the subthreshold region can be given as

(2.31)

The current expression can be obtained by integrating Eq.(2.31) along the channel from

source to drain,

(2.32)

where Qdinv and Qsinv are the channel inversion charge at the drain and source.

The channel charges at the source and drain can be written as,

(2.33)

(2.34)

Therefore, the current expression becomes

(2.35)

(2.36)

vt is the thermal voltage (KBT/q) and VOFF is the offset.

27
2.3.7 Other Important Parameters

1. The VOFF Parameter

The theoretical threshold voltage Vth,sub needed to fit the subthreshold current is different

form Vth that is used to fit strong inversion I-V. One explanation is that the surface

potential corresponding to the Vth in strong inversion is actually higher than 2φB. The

difference between the threshold voltages discussed above is several vt. To account for

this fact, a parameter called VOFF is introduced so that

(2.37)

VOFF is determined experimentally from the measured I-V characteristics and is expected

to be negative. Due to the physical meaning of VOFF, overly large absolute values of

VOFF are not recommended in the model. The recommended range for VOFF is

between -0.06 and -0.12V.

2. Drain and Source Parasitic Resistance, Rds

In BSIM3v3, the parasitic resistance of drain and source is modeld with th e following

expression:

(2.38)

where WR is a fitting parameter and RDSW has the units of Ω-μmWR. PRWB is the

body effect coefficient, and PRWG is the gate-bias effect coefficient.

3. The n Parameter for Subthreshold Swing

28
The n parameter can be called the subthreshold swing factor or the subthreshold slope

factor because the traditional gate voltage swing of subthreshold slope can be defined as

(2.39)

The subthreshold swing is the change in the gate voltage Vgs required to reduce the

subthreshold current Igs by one decade. Accordeing to Eq.(2.39), the n paramter is the

key paramter in determining the subthreshold swing of the device. For long channel

devices, n can be modeled as

(2.40)

where Cdep and Cit are the deplition layer capacitance and interface charge capacitance.

However, Eq.(2.40) does not consider the influence of short cahnnel effects. In short

channel devices, the potential at the surface of the channel will be determinde by both

the gate bias and the drain bias through the coupling of Cox and Cdsc, respectively, instead

of the gate bias only. The coupling capacitance Cdsc(L) is an exponential function of the

channel length. To reflect this phenomenon in BSIM3v3, the n parameter for the

subthreshold swing is described in the following form:

(2.41)

2.3.8 Source / Drain Series Resistance

The extrinsic parasitic resistances are important parameters which influence MOSFET

performance largely. In deep sub-micron device, it becomes much more important

29
because the parasitic resistances are independence even if scaled down.

Ideally, in the discussion of MOSFET current, we consider the source and drain regions

as conductor. However, in reality, as the current flows from the channel to the terminal

contact, there is a small voltage drop in the source and drain regions due to the finite

silicon resistivity and metal contact resistance. In a long-channel device, the source and

drain parasitic resistances are negligible compared with the channel resistance. In a

short-channel device, however, the source and drain series resistances can be an

appreciable fraction of the channel resistance and can therefore cause significant current

degradation. To model these parasitic resistances accurately, there are two approaches

to model the parasitic source and drain resistances in BSIM3: (a) lumped-resistance

approach and (b) absorbedresistance approach. It shows in Figure .

Figure 2.5 The parasitic source and drain resistances: (a) lumped-resistance approach
and (b) absorbed-resistance approach.

The lumped-resistance approach, while being conceptually simple and, in fact, more

accurate than the to-be-discussed second approach, has some drawbacks. First, there are

30
two more nodal voltages which need to be solved for each transistor; namely, the

Source’ and the Drain’ nodes. This translates to lengthened computation time in a

SPICE simulation. For simple circuit with few transistors, however, we believe this

extra amount of time needed for numerical solution is trivial. Second, generally SPICE

parameters are extracted by matching the measured and simulated values of the drain

current and the mutual transconductance. In a measurement setting, the measured

current is that at the Drain node, and the measured gm, for example, is equal to

@Id=@Vgs(still referring to Figure 2.5.(a)). The calculated drain current, which is the

current at the Drain’ node, will be the same current that flows through the Drain node.

Therefore, extracting parameters by fitting the simulated current at the Drain’ node

against the measured current at the Drain node makes sense. However, the calculated gm

from BSIM3 is really @I0d=@V 0 gs, which often differs significantly from measured

gm equaling to @Id=@Vgs. Hence, unless a numerically intensive SPICE subroutines

are incorporated to calculate the derivatives at the Drain’ and

Source’ nodes, SPICE parameters should not extracted by fitting the measured and

simulated g’ms.

Some formulas are proposed to correct this problem. With a certain small-signal

approximation, it is possible to relate the measured conductances to intrinsic

conductances:

(2.42)

(2.43)

(2.44)

31
where, for example, gd is @Id=@Vds, the measured drain conductance, and gd,i is the

intrinsic device conductance calculated in BSIM3, equal to @Id=@V 0 ds. After the

application of Eq.(2.43), it is possible to extract parameters by fitting BSIM3 generated

gd,i to the measured gd. We caution that, although the approximation of gd given in

Eq.(2.43) is fairly good, using Eq.(2.42) to relate gm and gm,i can run into problems

unless the current is fairly low. The equations expressed in Eqs.(2.42)-(2.44), after all,

are derived with the small-signal assumption, which can sometimes fail at normal

operating current levels.

BSIM3 offers another approach to model the parasitic source/drain resistances,

basically by absorbing the parasitic resistances’ effects on current and conductances into

the intrinsic device. The equivalent circuit is shown in the right-hand side of Figure 2.5.

In this absorbed-resistance approach, the nodes Drain’ and Source’ disappear. The drain

current and mutual transconductance are calculated from a totally different set of

equation

(2.45)

We use f 2 to describe the function form of ID,2, in order to stress that the function differs

from f 1 used in the calculation of ID,1. Basically, BSIM3 devises some sort of function f 2,

which is a modified version of f 1 to account for the effects of the parasitic resistances.

The amount of deviation from the original f 1 depends on the magnitude of RD+RS. The

32
larger the value of RD+RS, the more significant the deviation. The modification is made

in ways such that somehow ID,2 is approximately equal to ID,1 and more importantly, gd,2,

gm,2 and gmb,2 are approximately equal to gd,1, gm,1 and gmb,1, respectively. Note the keyword

above is approximately. No matter how clever the BSIM3 modification may be, there is

no way to replace the SPICE simulator’s mathematical calculations of the two

additional nodal voltages if the extract solution is desired. The approximation is most

accurate when the drain current is small such that the voltage drop across RS is small

compared to the applied VGS. However, despite some claims to the contrary, it is

believed that the error can be quite high (exceeding 10 ％ at least) when the current

flow is large. Setting the issues of dc accuracy aside, the absorbed-resistance approach

is also discouraged for RF applications. When the second approach is used, the input

resistance of the MOSFET will be purely imaginary, without seeing the effect of RS.

This is because BSIM3 makes an attempt to equilibrate only the drain current and the

conductances at the two sides of Figure 2.5. The fact that the device y-parameters of the

overall device are greatly altered when two extra nodes are added is not considered.

Here are some details of the actual BSIM3 implementation. When the lumped-resistance

approach is used, RS and RD are specified with the SPICE parameter RSH (sheet

resistance of the source and drain contact), together with the fields NRD (number of

squares in the drain contact) and NRS (number of squares in the source contact) found in

the device statement. They are given by

(2.46)

RDSW, to be discussed shortly, should be zero in this lumped-resistance approach.

In the absorbed-resistance approach, we see that the function f 2 in Eq.(2.45) depends

33
on the sum of RS and RD, rather than each component individually. From this

observation alone, we can infer that BSIM3 uses a completely new parameter for this

approach, independent from the aforementioned RSH, NRS and NRD. It is RDSW, the

sum of source and drain resistances per unit width. When this approach is used, RSH

should be set to zero, and RDSW can be approximated from the measured RS and RD:

(2.47)

where W is the device’s width. (The actual relationship between the parasitic resistances

and RDSW includes other parameters such as WR, PRWG and PRWB to account for

secondary effects. Eq.(2.47) is valid when WF, PRWG, and PRWB assume their default

values of 1, 0, and 0, respectively.) For a RF circuit designer who would like to see

nonzero real part input resitance due to the source resistance, the lumped-resistance

approach is preferred and RDSW should be zero.

2.4 Modeling of the Extrinsic MOSFET

Although many MOSFET models, including MOS9, EKV and BSIM3v3, have been

developed for digital, analog and mixed-signal applications, the RF MOSFET modeling

accuracy of the existing MOSFET compact model is not satisfactory. It is because, at

high frequency region, the parasitic components severely influence to device RF

performance and it is difficult to predict these parasitics behavior only within a compact

model. To make models more accurate at RF, sub-circuits are usually added to intrinsic

transistor model, BSIM3v3 in this paper, as shown in Figure 2.6. Simple sub-circuits are

preferred to reduce the simulation time and to make parameter extraction easier. For an

34
AC small signal model at RF, the modeling of sub-circuits components, i.e. the parasitic

components, is very important. The models for these parasitic components should be

physics-based and linked to process and geometry information to ensure the scalability

and prediction capabilities of the model. In this section, the methodologies of the

extrinsic parameter extraction, which surround the intrinsic MOSFET in Figure 2.6 will

be described.

Figure 2.6 Sub-circuit model with intrinsic and extrinsic components [12]

2.4.1 Gate Resistance Model

The typical sheet resistance for a poly-silicon gate ranges between 20-40 Ω/square and

can be reduced by a factor of 10 with a silicide process, and even more with a metal

stack process. At DC and low frequency, the gate resistance consists mainly of the

poly-silicon sheet resistance and is independent of bias. However, at high frequency, it

becomes a bias-dependent component and two additional physical effects will affect the

35
effective gate resistance [19][20]: the distributed gate electrode resistance (Reltd or Rpoly)

and non-quasi-static (NQS) effects (Rch), as shown in Figure 2.7. [14],

(2.48)

At first, the distributed gate electrode resistance (Reltd) will be described. It will become

more severe as the gate width becomes wider at higher operation frequency. So

multi-finger devices are used in the circuit design with narrow gate width for each

finger to reduce the influence of

Figure 2.7 Distributed gate electrode resistance Reltd, channel resistance Rch and gate
capacitance Cgg.

this effect. With multi-finger devices, gate resistance can separate, as shown in Figure

2.8 [19]. As Reltd is insensitive to bias and freqency, its value can be obtained from the

gate electrode sheet resistance (Reltd-sh),

(2.49)

(2.50)

where α is 1/3 when the gate terminal is brought out from one side, or 1/12 when

36
connected on both sides [21]. The value of α accounts for the distributed nature of the

RC line across the channel.

Figure 2.8 Divided Reltd into Rext and Rpoly with multi-finger devices.

Next, NQS effects (Rch) in the channel will be described. For the devices with NQS

effects, additional bias and geometry dependences of the gate resistance are needed to

account for the NQS effects. There are two mechanisms involved in Rch: one is the static

channel resistance (Rst), which accounts for the dc channel resistance and the other is the

excess diffusion channel resistance (Red) due to the change of channel charge

distribution by ac excitation of the gate voltage. Rst and Red together determine the time

constant of the NQS effects. Rst is modeled by integrating the resistance along the

channel under quasi-static assumption,

(2.51)

(2.52)

37
where Vdsat is the saturation drain voltage. Both Id and Vdsat are available in BSIM3v3. Red

can be derived from the diffusion current as

(2.53)

whereηis a technology-dependent constant. The overall

channel resistance seen from the gate is

(2.54)

where γ is a paremeter accounting for the distributed nature of the channel resistance

and Cox is oxide capacitance per unit area. γ equals to 12 if the resistance is uniformly

distributed along the channel. Since this assumption is not valid in the saturation region,

γ is left as a fitting parameter.

38
Chapter 3
RF Characteristics of High-k MOSFET

In this chapter RF characteristics of next generation sub-100nm high-k MOSFET are

discussed as being related to DC characteristics and device structure. RF characteristics

of high-k MOSFET is analyzed based mainly on gate resistance and capacitance

extracted from measured S-parameter up to 40 GHz. In consequence RF application of

high-k MOSFET is explored by examining various issues of high-k MOSFET.

3.1 Device structure

In this chapter two type of high-k MOSFET which have higher dielectric constant

gate insulator than SiO2; one is HfSiON and the other is SiON, were focused on. High-k

technology is now researched and developed actively in device process research. There

are a lot of papers about process technology on high-k. With down scaling of device

gate leakage current has been significant problem because it leads to high electricity

dissipation. High-k prevents gate leakage current caused by electron tunneling at gate

insulator increasing. However, there are some concerns on high-k MOSFET. First is

degradation of channel electron mobility by surface roughness. This may affect to RF

characteristics of device. Second is fall of dielectric constant in RF region by dielectric

dispersion. If this is truth high gate capacitance with a thick gate insulator, one of merits

of high-k MOSFET, can not be kept in RF region. The merit of high-k MOSFET is lost.

39
 Silicide
Silicide
HfSiON SiN SiON SiN

Si Si

（a) HfSiON (b) SiON

Figure 3.1 Transistor structure TEM

Cross section of transistor structure is indicated at Figure 3.1 and (a), (b) is HfSiON

device and SiON device respectively (HfSiON device and SiON device is mentioned as

HfSiON and SiON simply at the following). High-k gate insulator pile up on Si

substrate and Silicide is done at electrodes. EOT is 1.5 [nm] on both device.

In this study multi gate finger structure is applied to reduce gate resistance. Gate

resistance Rg is described by Eq. 3.1.

1 Wf 1 W
R g , eff = Rsh = Rsh total2 (3.1)
3 LN f 3 LN f

Wf: Finger length, Nf: Number of finger

As Eq. (3.1) indicates, as number of gate finger increase, Rg is reduced. Plane Figure is

indicated at Figure 3.2. Unit device which is indicated at Figure 3.1 is arranged in

parallel.

40
 Gate finger
G

Finger length
S
S G S

D
S D S

M1
G G G G G
VIA1

STI S D S D S

Figure 3.2 Multi gate finger structure

41
 3.2 DC Characteristics

Figure 3.3 shows DC characteristics of high-k MOSFET. One is HfSiON device and
another is SiON device. Both devices are fabricated in 90 nm technology node.

Effective gate length of HfSiON device is 64 [nm] and that of SiON device is 51 [nm].

Number of gate fingers is 12 and finger width is 5 [um], this is equal to that gate width

is 60 [um].

0.03 0.03

0.025 SiON 0.025

SiON
Id[A]

Id[A]

0.02
HfSiON 0.02
0.015 0.015

0.01 0.01 HfSiON

0.005 0.005
0 0
0 0.4 0.8 1.2 0 0.4 0.8 1.2
Vd (V) Vg (V)
(a) Id-Vg characteristics (b) gm-Vg characteristics
@ Vg = 0.6, 0.8, 1.0, 1.2, 1.5 @Vd = 0.6, 0.8, 1.0, 1.2 ,1.5

0 .0 5

0 .0 4 SiON
gm

0 .0 3

0 .0 2

0 .0 1
HfSiON
0
0 0 .4 0 .8 1 .2
Vg (V)
(c)Id-Vd charcteristics

Figure 3.3 DC characteristics

42
As this figure indicates, SiON device has better DC characteristics. It seems that this is

due to the difference of gate length and electron mobility at channel. At Figure 3.3 (c),

SiON device of transconductance gm (Eq. (3.2)) is higher than that of HfSiON device.

gm is affected by gate length, gate width and mobility. This difference of gm affects RF

characteristics as mention later.

∂Id W
gm = = μC oxVd (3.2)
∂Vg L

Figure 3.4 shows DC characteristics at different temperature condition. DC

measurement was performed at various conditions in addition to room temperature.

Plot Hfno3/L090W 5F12001L/dc_idvg/idvg PO Plot Hfno3/L090W5F12001L/dc_idvg/gmvg
id.m L090W5F12001/dc_idvg/id.m L090W5F12001H/dc_idvg/id.m [E-3]

gm.m L090W5F12001/dc_idvg/gm.m L090W5F12001H/dc_idvg/gm.m [E-3]

25 40
A

20
gm [mS]

30
Id [mA]

15

20

10

10
5

0 X 0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 0.0 0.2 0.4 0.6 0.8 1.0 1.2
vg [V]
Vg [E+0] Y vg [E+0]
Plot Hfno3/L090W5F12001L/dc_idvd/idvd
Vg [V]
PO
id.m L090W5F12001/dc_idvd/id.m L090W5F12001H/dc_idvd/id.m [E-3]

25
A
-40℃
20
Id [mA]

15 RT

10
105℃
5

0 X
0.0 0.2 0.4 0.6 0.8 1.0 1.2
vd [E+0] Y
Vd [V]

Figure 3.4 Temperature dependency of DC characteristics of HfSiON device

43
Trans conductance gm improved at -40 oC. Mobility improvement at low temperature

leads to gm improvement.

44
3.3 RF Characteristics

3.3.1 Cutoff Frequency and Maximum Oscillation Frequency

S-parameter were measured up to 40GHz and cutoff frequency fT and maximum

oscillation frequency fmax are extracted from measured S-parameter. They are important

figure of RF characteristics. Figure 3.5 indicates H21 which is 21 entry of H-parameter

of device and Figure 3.6 indicates Unilateral gain which is described by Eq. (1.2). Those

are gotten from measured S-parameter.

[dB]
25
H21
20
HfSiON
SiON
15

10

0
10 100
1.0E+09 1.0E+10 [GHz]1.0E+11
Frequency
Figure 3.5 H21 @ gm peak bias; Vg = 1.2 [V], Vd = 1.2 [V]

45
 [dB]
90
80 Unilateral gain
70
60
HfSiON
50 SiON
40
30
20
10
0
10 100
1.0E+09 1.0E+10 [GHz]1.0E+11
Frequency

Figure 3.6 Unilateral gain @ gm peak biasVg = 1.2 [V], Vd = 1.2 [V]

fT and fmax are the frequency at which H21 and unilateral gain is 0 [dB]. The value of fT

and fmax extracted from Figure 3.5 and 3.6 is shown at Table 3.1.

Table 3.1 fT and fmax @ 5 [um] finger

HfSiON SiON

fT [GHz] 131 155

fmax [GHz] 58 53

As for fT, the value over 100 GHz were gotten on both devices. However, there are

difference between two devices, fT of SiON device is bigger than that of HfSiON device.

This is due to the difference on gm on DC characteristics shown in Figure 3.3. gm of

46
SiON device is bigger than that of HfSiON device. Gate voltage dependency of gm and

fT are shown in Figure 3.7.

160 70
HfSiON
140 60
SiON
120
50
fT [GHz]

gｍ [mS]
100
40
80
30
60
20
40
20 10

0 0
0.1 1 10 100
Id [mA]

Figure 3.7 Gate voltage dependency of gm and fT @ Vd = 1.2

fT reflects to the transition of gm, depending on device structure and mobility

As for fmax, extracted values were low and there are little difference between HfSiON

and SiON. It is known that fmax is not scaled. However, this low fmax is affected by gate

resistance. Extracted gate resistance from measured S-parameter is shown in Figure3.8.

47
 100

90 HfSiON
Gate resistance [Ω]
80 SiON
70

60

50

40

30

20

10

0
1 0 0 010
0000000 1100
E+ 1 1
Frequency [GHz]

Figure 3.8 Extracted gate resistance Rg

As Figure 3.8 shows, gate resistance Rg is large. fmax is described by Eq. (1.3).

Therefore large Rg leads to low fmax. And Figure 3.8 shows that Rg of SiON device is

bigger than that of HfSiON device. Although there are a little difference on gate length

among two devices, it can be thought that fmax of HfSiON device is bigger than that of

SiON device because Rg of SiON device is bigger. Gate resistance Rg is analyzed later.

Temperature dependent of fT, fmax is shown in Figure 3.9. fT and fmax is reflected by

temperature dependent of DC characteristics. Mobility improvement at low temperature

affect RF characteristics.

48
 fT

200
HfSiON @ - 40 oC
180
HfSiON @ 105 oC
160
SiON @ - 40 oC
fT [GHz]

140 SiON @ 105 oC

120
100
80
60
40
20
0
0.01 0.1 1 10 100
Id [mA]

(a) Temperature dependent of fT

fmax

70
HfSiON @ - 40 oC
60 HfSiON @ 105 oC
SiON @ - 40 oC
50
fmax [GHz]

SiON @ 105 oC
40

30

20

10

0
0.01 0.1 1 10 100
Id [mA]
(b) Temperature dependent of fmax

Figure 3.9 Temperature dependent of fT and fmax

49
On the other hand, device which has 2 [um] of finger length but the other structure is

the same were also analyzed. H21 and unilateral gain of the device are shown in Figure

3.10 and 3.11. Figure 3.10 shows H21 and unilateral gain of HfSiON device, Figure 3.11

shows those of SiON device.

[dB]
50
45 H21
40 Unilateral gain
35
30
25
20
15
10
5
0
1.00E+08 1
1.00E+09 10
1.00E+10 100
1.00E+11

Frequency [GHz]

Figure 3.10 H21 and unilateral gain of HfSiON

50
 [dB]
50
H21
45
Unilateral gain
40
35
30
25
20
15
10
5
0
1.00E+08 1
1.00E+09 10
1.00E+10 100
1.00E+11
Frequency [GHz]

Figure 3.11 H21 and unilateral gain of SiON

fT and fmax from extending H21 and unilateral gain at Figure 3.10 and 3.11 are shown in

Table3.2.

Table 3.2 fT and fmax @ 2 [um] finger

HfSiON SiON

fT [GHz] 147.91 177.83

fmax [GHz] 121.62 127.35

As Table 3.2 shows, both fT and fmax were improved at 2 [um] finger in comparison with

5 [um] finger. This corresponds with simulation result on gate finger length dependency

of RF figure (fT and fmax) which is indicated in [5]. Gate finger length has a optimum

point. RF MOSFET has best performance in terms of fT and fmax at the optimum point.

51
3.4 Parameter Extraction

3.4.1 Extraction of DC Parameter

Extraction of DC parameter by fitting using BSIM 3 was performed on HfSiON

device and SiON devise. Both device have same structure; gate length is 58.6 [nm],

number of finger is 48, finger length is 2 [um] (if not otherwise specified the same

structure is applied in what follows). Measured DC data and simulation data of HfSiON

by BSIM4 is shown in figure 3.12 and those of SiON is shown in Figure 3.13.

6 A 10
Symbols: Measured Symbols: Measured
5 E
Solid line: Simulation 8 Solid line: Simulation
gm.m gm.s [E-3]

4
Id [mA]
id .m idId.s [mA]
[E -3 ]

6
3
4
2
2
1

0 0
0.0 0.5 1.0 1.5 X 0.0 0.5 1.0 1.5
Vg [V]
vg [E+0] Vg[E+0]
vg [V]
Y
PO
Plot kakuHf110/Hf110DW
(a) Id-Vg characteristics @ Vd = 0.05V 2F48/dc_idvd/idvd
(b) gm-Vg characteristics @ Vd = 0.05V
A
60 E
Symbols: Measured
50
Solid line: Simulation
[E-3]

40
Id [mA]

30
id.m id.s

20
10
0
0.0 0.5 1.0 1.5
X
vdVd[E+0]
[V]
Y
(c) Id-Vd characteristics @ Vg=0.25, 0.50, 0.75, 1.0, 1.25, 1.5

Figure 3.12 DC modeling of HfSiON device

52
 Plot kakuSi110/Si110DW 2F48/dc_idvg/gmvg PO
Plot kakuSi110/Si110DW2F48/dc_idvg/idvg PO
A
8 A 15
Symbols: Measured E Symbols: Measured E

gm .m gm .s [E -3]
6 Solid line: Simulation Solid line: Simulation
10

g m [mS]
Id [mA]
[E-3]

4
id.m id.s

5
2

0 0
0.0 0.5 1.0 1.5 X
0.0 0.5 1.0 1.5
vgVg[E+0]
[V] Y Vg [V] X
(a) Id-Vg characteristics @ Vd=0.05V (b) gm -Vgvg [E+0] @ Vd=0.05V
characteristics Y
Plot kakuSi110/Si110DW 2F48/dc_idvd/idvd PO
100 A
Symbols: Measured
E
80
Solid line: Simulation
Id [mA]
id.m id.s [E-3]

60

40

20

0
0.0 0.5 1.0 1.5 X
vd
Vd[E+0]
[V] Y
(c) Id-Vd characteristics @ Vg=0.25, 0.5, 0.75, 1.0, 1.25, 1.5 V

Figure 3.13 DC modeling of SiON device

Simulation data fits measured data well on both devices. Principal difference between

two devices is threshold voltage. Vth of HfSiON is 0.7262 [V], Vth of SiON is 0.4582

[V]. As for channel mobility, there are a little difference between two devices. μ0 of

HfSiON is 356.5 [cm2/V] and μ0 of SiON is 319.2 [cm2/V].

53
3.4.2 Extraction of Extrinsic Components by Using

Equivalent Circuit

In this section, terminal resistances (gate resistance Rg, drain resistance Rs, source
resistance Rs) are focused on. The extraction of the terminal resistances is done by

performing Z-parameter analysis on the equivalent circuit at the linear region of the

transistor. It is proposed that the extraction of the resistances is done at Vg=1.5 V and

Vd = 0V. At this bias condition, the transconductance and transcapacitance are both

close to zero and influence by substrate components can be neglected[13]. The

simplified equivalent circuit of MOSFET at Vg=1.5 V and Vd=0 V is shown in

Figure3.14.

gox

Rg Rd
G D
Cgate
Rs

Figure 3.14 Equivalent circuit @ Vg =1.5 V and Vd=0V

54
The resistance Rg represents the effective lumped gate resistance, consisting of both the

electrode resistance and the distributed channel resistance [14]. It has been reported that

a simple gate resistance model has been found accurate up to 0.5 fT [15]. The resistance

Rd and Rs represent effective drain and source resistances that consist of the via,

salicide, contact and LDD resistances [16]. The capacitance Cgate is gate capacitance

consist of effective gate-to-drain capacitances and gate-to-source capacitances, intrinsic

gate insulator capacitance and gate overlap capacitance. gox represents the conductance

of gate insulator part, which means easiness of gate leakage current.

In Figure 3.12, Z-parameter can be written as below.

Rl
Z11 = Rg + Rs +
jωCgate ⋅ Rl + 1
Z12 = Z 21 = Rs (3.3)
Z 22 = Rd + Rs

At Eq. (3.3), Rl = 1 g ox . By using Eq.(3.3), components of equivalent circuit can be

extracted from measured S-parameter. In this chapter, analyzed device is finger length is

2 [um], number of finger is 48, gate length is sub-100nm.

55
3.4.3 Influence of Gate Leakage Current to Gate Resistance

Gate leakage current of MOSFET is important factor in sub-100 nm generation. It

causes increase of electrical dissipation. This is a motivation for adopting high-k

MOSFET. Although two type of devices which have HfSiON and SiON as gate

insulator are interest in this study, dielectric constant of HfSiON is larger than that of

SiON. Therefore, gate leakage current of SiON is lareger than that of HfSiON. In this

section, influence of gate leakage current to gate resistance is discussed. As mentioned

previously, gate resistance is important parameter of RF MOSFET.

Gate leakage current is shown in Figure 3.15.

Plot kakuHf110/Hf110DW 2F48/dc_igvg/idvg PO

1E-7
abs(ig.m) abs(kakuSi110/Si110DW2F48/dc_igvg/ig.m) [LOG]

A
HfSiON
1E-8
SiON

1E-9
Ig [A]

1E-10

1E-11

1E-12

1E-13 X
-2 -1 0 1 2
vg.mVg [V]
[E+0] Y

Figure 3.15. Gate leakage current of two devices

There is difference of gate leakage current, leakage current of SiON is larger than that

56
of HfSiON. This means that Rl ( =1/gox ) of HfSiON is larger than that of SiON. At

Vg=1.5 V, the difference of gate leakage current between two devices is two columns.

In this section, Z-parameter analysis is performed. Real part of Z11 is described at Eq.

(3.4) based on Figure 3.14.

Rl
real ( Z11) = Rg + Rs +
1 + (ωCgateRl ) 2
(3.4),
ωCgateRl 2
imag ( Z11) = −
1 + (ωCgateRl ) 2
Real part of Z11 and imaginary part of Z11 calculated from measured S-parameter up to

40GHz are shown in Figure 3.16 and 3.17.

30

25
HfSiON
Real (Z11)

20 SiON

15

10

0
10
1E+10 1.5E+10 20
2E+10 2.5E+10 30
3E+10 3.5E+10 40
4E+10

Frequency [GHz]
Figure 3.16 Real part of Z11 at Vg = 1.5 V

57
 0
-50
-100
-150
Imag (Z11)

-200
-250
HfSiON
-300
SiON
-350
-400
-450
-500
1E+10
10 1.5E+10 2E+10
20 2.5E+10 30
3E+10 3.5E+10 40
4E+10

Frequency [GHz]
Figure 3.17 Frequency dependent of imaginary part of Z11

As for real(Z11) and imag(Z11), there are little difference between two devices.

Real part of Z12 which represents Rs by Eq. (3.3) is shown in Figure 3.18.

10
9
8 HfSiON
7 SiON
Rs [Ω]

6
5
4
3
2
1
0
1.00E+10
10 1.50E+10 2.00E+10
20 30
2.50E+10 3.00E+10 40
3.50E+10 4.00E+10

Frequency [GHz]

Figure 3.18 Real part of Z12 which represents Rs

58
There are a little difference between two devices on Rs.

Next, Rg is extracted from measured S-parameter. By using Eq. (3.3) and (3.4),
Rl
real ( Z11) − real ( Z12) = Rg + (3.5).
1 + (ωCgateRl ) 2

At high frequency region, second term of Eq. (3.5) can be neglected. Therefore Eq. (3.5)

represents Rg. By using real(Z11) shown in Figure 3.16 and real(Z12) shown in Figure

3.18, real part of series parameter of gate part “real(Z11)-real(Z12)” is shown in Figure

3.19.

25
Real(Z11) – real(Z12)≒Rg

HfSiON
20
SiON

15

10

0
10
1.00E+10 1.50E+10 2.00E+10 40
30 3.50E+10 4.00E+10
20 2.50E+10 3.00E+10
Frequency [GHz]
Figure 3.19 Real part of series parameter of Gate part

real(Z11)-real(Z12) is little different from real(Z11) because real(Z12) is small in

comparison with real(Z11)-real(Z12). And the difference between two devices on

real(Z11)-real(Z12) can be seen.

59
Then, ral(Z11)-real(Z12) from measured S-parameter compare with calculated

real(Z11)-real(Z12) and discuss if Rl can affect Rg.

How to calculate real(Z11)-real(Z12) is indicated as below. At first, gox can be extract

by using Ig-Vg characteristics of MOSFET, Figure 3.15. gox represents Ig divided by

Vg. Ig is 664.1 [pA] on HfSiON, 60100 [pA] on SiON at Vg=1.5 V, Vd=0 V. gox at the

bias is shown in Table 3.3.

Table 3.3 Gate Leakage Component gox on both devices

HfSiON SiON

gox 4.43*10-10 4.01*10-8

By using Rg extracted from measured S-parameter and gox from DC characteristics,

Cgate from device structure, real(Z11)-real(Z12) can be calculated. It is shown in

Figure 3.20 where the calculated value of real(Z11)-real(Z12) when gate leakage

current increase are plotted altogether.

25
Calculated @ Ig= 0.665 [mA]
real(Z11)-real(Z12)≒Rg

20 Calculated @ Ig= 0.0665 [mA]

Measured
15

10

0
10 1.50E+10 2.00E+10
1.00E+10 20 30
2.50E+10 3.00E+10 40
3.50E+10 4.00E+10
Frequency [GHz]

Figure 3.20 Calculated Rg at gate leakage current Ig variety

60
As figure 3.20 shows, gate leakage component can not be neglected when Ig is 10-1 mA

order. Moreover the influence is bigger at low frequency region which is under 30 GHz

in Figure 3.20. And a little increase of gate leakage current can not be affect to

“real(Z11)-real(Z12)”. Therefore, the difference of gate leakage current between two

devices can not affect to Rg.

3.4.4 fT and fmax in Conjunction with Gate Resistance

Three devices that gate lengths are different (37.8nm, 51.3nm, 58.9nm) were measured.
The three devices have same structure, only gate length is different. Number of gate

finger is 48, finger length is 2 [um]. EOT is 1.5 [nm]. Gate length dependency of fT and

fmax are shown in Figure 3.21.

[GHz]
250
fT
200 fmax
150

100

50

0
10 100
Gate length [nm]

Figure 3.21 Gate length dependency of fT and fmax

61
fT is scaled with gate length although fmax is not scaled with gate length. This can be

thought due to Rg. Gate length dependency of Rg is shown in Figure 3.22.

25
37.8 [nm]
20 51.3 [nm]
58.6 [nm]
Rg [Ω]

15

10

0
10
1.E+10 20
2.E+10 30
3.E+10 40
4.E+10
Frequency [GHz]

Figure 3.22 Gate length dependency of gate resistance

Although gate resistance Rg is frequency dependent in figure 3.22, trend of gate length

dependency of Rg can be seen. Rg of 37.8 nm device is larger than those of the other

two devices. It can be said that Rg increases as gate length shrinks. Therefore scaling of

gate length does not always cause improvement of fmax although fT is scaled.

3.4.5 Analysis of Gate Capacitance

Gate capacitance is discussed in this section. As mentioned previously, there is a

concern that gate capacitance may decline by dielectric dispersion at RF region.

62
Therefore gate capacitance extracted measured S-parameter is analyzed. How to extract

gate capacitance is described as below.

Gate capacitance was extracted at Vg = 1.5 V and Vd = 0 V. At this bias condition,

simplified equivalent circuit which represents Z11 is shown in Figure 3.23.

Rg

Cgate gox

Rseries
∋ Rs, Rd

Figure 3.23 Simplified equivalent circuit

Therefore Z11 can be described by Eq. (3.6).

1
Z11 = Rg + Rseries + (3.6)
jωCgate + g ox

Then Cgate can be extracted by de-embedding Rg and Rseries at Eq. (3.7),

imaginary{( Z11 − Rg − Rseries) −1 }

Cgate = (3.7)
ω

63
At Eq. (3.3), supposing gox can be neglected,

Rg = real ( Z11) − real ( Z12)

(3.8)
Rseries = Rd + Rs = real ( Z 22)

Extracted gate capacitance based on Eq. (3.7) is shown in Figure 3.24.

100
95 HfSiON
90
Capacitance [fF]

SiON
85
80
75
70
65
60
55
50
-2 -1 0 1 2
Vg [V]

Figure 3.24 C-V characteristics @ 10GHz

As mentioned at DC modeling section, there is difference of threshold voltage between

two devices. As Figure 3.24 indicates, the difference on Vth can be seen. The leading

edge of capacitance corresponds to the Vth on both devices. Moreover capacitances at

strong inversion region almost fit in because two devices have same EOT.

Extracted capacitance value up to 40 GHz is shown in Figure 3.25.

64
 200

175 HfSiON
Capacitance [fF]

150 SiON

125

100

75

50
0 10 20 30 40
Frequency [GHz]

Figure 3.25 Extracted capacitance at wide-ranging frequency

As Figure 3.25 indicates, capacitance value is constant up to 40 GHz on both devices. It

means that capacitance degradation caused by dielectric dispersion does not arise up to

40 GHz. At this point of view, the concern which high-k is not high-k at RF region is

denied.

65
Chapter 4
Exploration for Optimum Device Structure

In this chapter, minimum noise figure and distortion behavior of conventional

MOSFET which have silicon oxidation membrane are mainly analyzed. What is

focused on is device structure which means gate finger length and how to arrange gate

finger. In consequence, optimum device layout is examined.

4.1 Device structure

Figure 4.1 shows plane figure of analyzed device.

Finger length

Module: diffusion area

Finger: Number of
gate array on uniform
diffusion area

(a) Plane view

G G G G G

S D S D S D
Lg Lg Lg Lg Lg

(b) Cross section

Figure 4.1 Device strucutre

66
Module means diffusion area; number of module is 4 in figure 4.1. Finger and finger

length is same definition in chapter 3; number of finger represents number of gate array

on uniform diffusion area, finger length represents unit gate array length. Gate insulator

is conventional silicon oxidation membrane. Multi gate finger device has various

parameters as just described. Among various parameters, finger length is focused on. As

finger length is varied, RF analysis is performed.

Finger length variety is shown in Table 3.4. Gate length is 0.18 [um], gate width is

constant at 160 [um] on all devices.

Table 3.4 Device List

Wf [um] Number of Number of
Finger Module
1 40 4

1.25 32 4

2 20 4

2.5 16 4

4 10 4

5 8 4

8 5 4

10 4 4

20 2 4

4.2 DC characteristics

Dc characteristics when finger length Wf is varied are shown in Figure 4.2.

67
 __ _ __ _ g g__
A
id.m L018F16M4_Wf1p25/dc_idvg/id.m L018F16M4_Wf10/dc_idvg/id.m L018F16M4_Wf20/dc_idvg/id.m [E-3] 80
Wf = 4 [um]
60
Id [mA]

40

20

0
0.0 0.5 1.0 1.5 X
Vg [V]
vg [E+0] Y
(a) Id-Vg characteristics at various Wf

A
80
m L018F16M4_Wf1p25/dc_idvd/id.m L018F16M4_Wf10/dc_idvd/id.m L018F16M4_Wf20/dc_idvd/id.m [E-3]

Wf = 4 [um]

60
Id [mA]

40

20

0
0.0 0.5 1.0 1.5 X
Vｄ [V]
vd [E+0] Y
(b) Id-Vd characteristics at various finger length

Figure 4.2 DC characteristics at various Wf

68
It can be seen that there are a little difference among each DC characteristics at various

finger length although total gate width is same. This is due to the difference on wiring

resistance. At the assumption that gate width is constant, as finger length Wf increases

wiring resistance decreases as shown in Figure 4.3.

Gate Wiring resistance: small

Wiring resistance: large

Drain

(a) Small finger length (b) Large finger length

Figure 4.3 Relation with wiring resistance and finger length from plane view

Therefore finger length Wf, by decreasing which gate resistance can be decreased,

trades off with wiring resistance which represents the resistance from gate pad to gate

finger. This means that there is optimum finger length. In figure 4.2, DC characteristics

at Wf = 4[um] has largest drain current among 9 devices.

Actually, resistance (Rd, Rs) in array is extracted from DC characteristics. How to

extract is fitting simulation data with measured data optimizing Rd and Rs which are

embedded to intrinsic BSIM model of MOSFET as sub circuit. Figure 4.4 shows the

concept of intrinsic MOSFET and Rd and Rs as extrinsic components.

69
 Intrinsic MOSFET
G

Rs Rd
S D

Figure 4.4 Illustration with intrinsic MOSFET and Rd and Rs

Simulation data by only intrinsic MOSFET model is shown in Figure 4.5. Moreover

optimized simulation data when Rd and Rs as sub-circuit components are embedded as

figure 4.4 shows is shown in Figure 4.6.

PO LS5205__1/L018F16M4_W f4__1/dc_idvd/idvd
Plot LS5205__1/L018F16M4_W f4__1/dc_idvg/idvg Plot
150 A
150
Symbols: Measured E Symbols: Measured
Solid lines: Simulation Solid lines: Simulation
id .m id .s [E -3 ]

id .m id .s [E -3 ]

100 100
Id [m A]

Id [m A]

50 50

0 0
0.0 0.5 1.0 1.5 X 0.0 0.5 1.0 1.5
Vg [V]
vg [E+0] Vd [V]
(a) Id – Vg characteristics Y (b) Id vd
– Vd[E+0]
characteristics

Figure 4.5 Intrinsic MOSFET model

70
 100 A 100
Symbols: Measured
Symbols: Measured E
80 80 Solid lines: Simulation
Solid lines: Simulation
]

id .m id .s [E -3 ]
60 60

Id [m A]
Id [m A]
[

40 40
20 20
0 0
0.0 0.5 1.0 1.5 X 0.0 0.5 1.0 1.5
Vg [V] Vd [V]
vg [E+0]
(a) Id - Vg characteristics Y (b) Id - Vd[E+0]
vd characteristics

Figure 4.6 After embedding extrinsic components

After embedding Rd and Rs simulation data fits well. In this way, the extracted value of

Rd and Rs by fitting to measured data is shown in Figure 4.7.

[Ω]
4 .5

3 .5 Rs
3

2 .5

2 Rd
1 .5

0 .5

0
1 10 100
Wf [um]

Figure 4.7 Finger length dependency of Extracted Rd and Rs

71
As Figure 4.7 indicates Rd and Rs have minimum point. The Value of Rd and Rs of the

device of Wf = 4 [um] which have best performance on DC characteristics among 9

devices is lower. On the contrary, those of the device of Wf = 20 [um] which have worst

performance among 9 devices is larger. At Vg=Vd=1.5 V, the drain current value of

Wf=20 um device deteriorated by 6.5 % in comparison with that of Wf =4 um device.

Figure 4.8 shows the comparison with best case and worst case.

A
80 80
id .m L 0 1 8 F 1 6 M 4 _ W f2 0 /d c_ id vd /id .m [E -3 ]
id .m L 0 1 8 F 1 6 M 4 _ W f2 0 /d c _ id v g /id .m [E -3 ]

Wf = 4 [um]; best performance

60 Wf = 20 [um]; worst performance 60

Id [m A]
Id [m A]

40 40

20 20

0 0
0.0 0.5 1.0 1.5 0.0 0.5 1.0 1.5
X
vg Vg[E+0]
[V] vdVd[E+0]
[V]
Y
(a) Id – Vg characteristics (b) Id – Vd characteristics

Figure 4.8 Comparison with best case and worst case

72
4.3 RF Characteristics

4.3.1 Finger Length Dependency of fT and fmax

fT and fmax when finger length are varied were extracted from measured S-parameter up
to 40 GHz. Figure 4.9 shows drain current dependency of fT and fmax when finger length

are varied.

55 120
50 Wf=1
Wf=1.25 100 Wf=1
45
Wf=2 Wf=1.25
40 Wf=2.5 80 Wf=2
fmax [GHz]
fT [GHz]

35 Wf=4 Wf=2.5
Wf=5 60 Wf=4
30
Wf=8 Wf=5
25 Wf=10 40 Wf=8
20 Wf=20 Wf=10
20 Wf=20
15

10 0
0 .1 1 10 100 0 .1 1 10 100
Id [mA] Id [mA]
(a) fT ; Vd = 1.5 V, Vg is varied (b) fmax ; Vd = 1.5 V, Vg is varied

Figure 4.9 Finger length dependency of fT and fmax

fT and fmax are fluctuant depending on Wf although total gate width is constant. As for fT,

when Wf = 1, 1.25 um the value is very low in comparison with the other devices. The

other values is not so fluctuant depending on Wf. As for fmax, when Wf = 20 um, the

value is very low and when Wf = 4 um, the value is largest. About 2.5 times

improvement was seen in the value of fmax of Wf=4 um device in comparison with that

of Wf=20 um device.

73
4.3.2 Noise Characteristics

The measured NFmin (minimum Noise Figure) and Gav (Available Gain) on Wf = 2.5
um at different bias condition is plotted in figure 4.10. Measurement frequency is 5

GHz and 10 GHz.

5 25
5GHz

GAv (Available gain)[dB]

20
4 10GHz
15
NFmin [dB]

3 10

5
2
0

1 -5
0.1 1 10 100
Id [mA]

Figure 4.10 Noise Figure and Available gain versus drain current

Bias dependency of NFmin and Gav can be seen. Minimum value of measured NFmin is

1.29 @5GHz, 1.66 @10GHz at Vg=0.631 V, Vd=1.5 V. Moreover figure 4.10 indicates

that as Gav increase, NFmin can be minimized. NFmin can be minimized when Gav

become maximum. To conquer noise, gain of device needs to be large.

74
NFmin and gate resistance Rg are plotted in Figure 4.11. Rg was extracted by using Rd

and Rs optimized at DC characteristics[12].

3 25

2 .5 20
2
NFmin [dB]

10GHｚ 15

Rg [Ω]
1 .5
10
1 5GHｚ

0 .5 5

0 0
1 10 100
Wf [um]

Figure 4.11 Gate length dependency of NFmin related with Rg

NFmin are also fluctuant depending on Wf. The value of NFmin at Wf=4 um is lowest.

With the object of fmax and NFmin, it can be said that Wf=4 um is optimum finger length.

And Figure 4.11 indicates that NFmin responds to Rg. Minimizing gate resistance leads

to not only fmax improvement but also NFmin improvement.

75
Chapter 5
Conclusions and Future Issue
5.1 Conclusion

In this study, RF performance of high-k MOSFET and finger length varied multi finger
MOSFET for next generation were analyzed based on extracted parameters.

As for high-k MOSFET, the fT and fmax over 100 GHz were gotten on both devices,

however, HfSiON device has lower fT and fmax in comparison with SiON device. As for

fT, it is due to the difference of DC characteristics, mainly transconductance gm. And

scaling of device geometry causes the improvement. As for fmax, on the other hand, the

difference of gate resistance affects to the fmax of two devices. And scaling does not

always cause the improvement. Therefore it can be said that improvement of gm and

minimizing gate resistance leads to the improvement of RF performance. Moreover, RF

analysis relating with gate leakage current was performed. In consequence, it was

confirmed that gate leakage current does not affect to RF performance although there

are 2 column difference on gate leakage current between HfSiON device and SiON

device. It also can be said that low gate leakage current which is the merit of high-k

device can not improve RF performance. However, low power consumption by the

merit is very significant in RF circuits. In addition, fitting at DC was performed.

Simulation data by BSIM fitted with measured data well. The values of mobility from

optimizing was over 300 [cm2/Vs] on both devices. Mobility degradation of high-k

MOSFET was seen so much on these devices. And gate capacitance analysis at RF

region was performed. In terms of the concern; degradation of insulator capacitance at

RF region by dielectric dispersion, it was confirmed that the capacitance was constant

76
even if at RF region. As noted above, high-k MOSFET has potential ability at RF

region.

As for discussion about gate finger structure, it was confirmed that Wf=4 um is

optimum length at total gate width 160 um from the perspective of DC performance and

fmax. This is due to the wiring resistance in gate array. In addition, NFmin also has

optimum point depending on finger length and the relation with gate resistance was able

to be seen.

5.2 Future Issue

As future issue, RF modeling needs to be performed. In this study DC modeling was

performed. However, high-accuracy RF modeling is necessity for RF circuit simulation

and RF circuit designing. Moreover, more accurate equivalent circuit is needed,

although simplified equivalent circuit was adopted in this study. By doing this

MOSFET performance can be figured out more precisely. And dielectric dispersion

does not occur up to 40 GHz at gate insulator with small-signal in this study, however,

large-signal analysis needs to be performed. As for Noise characteristics, more thorough

analysis is needed including noise de-embedding. These leads to more exactly

understanding for RF performance of next generation device and RF application of

scaled CMOS including high-k device.

77
References
[1] Semiconductor Industry Association, et al., “International Technology Roadmap for
Semiconductors” (ITRS),” 2003 Edition.

[2] P.R.Gray and R.G.Meyer, “Future Directions in Silicon IC for RF Personal

Communications,” in Proc.IEEE Custom Integrated Circuits Conference, pp. 83-90,
May, 1995.

[3] A.Abidi, “Low-Power Radio-Frequency IC’s for Portable Communications,” in

Proc.IEEE, vol.83, no.4, pp. 544-569, 1995.

[4] Semiconductor Industry Association, et al., “International Technology Roadmap for

Semiconductors (ITRS),” 2005 Edition.

[5] E.Morifuji, H.S.Momose, T.Ohguro, H.Iwai, et al., “Future perspective and scaling
downroadmap for RF CMOS,” Symp. VLSI Technology, pp.163-164, 1999.

[6] H.Sauddin, “Low-Frequency (1/f) Noise of NMISFET with La2O3 High-k Gate
Dielectrics and Analysis of Electrical Properties of Ultra-Shallow p+/n Junctions
Formed by Plasma Doping Method,” Master thesis, Feb, 2005.

[7]Y.Tominaga, “Fundamental of Dielectric Dispersion and Physics of Microwave

Domain” in http://atom11.phys.ocha.ac.jp/ftp/pub/lecture/dielectric_relaxation.pdf

[8] C. J. F. Bottcher and P. Bordewijk, “Theory of electric polarization,” vol. II.

Elsevier, second ed., 1978.

[9] M.C.A.M.Koolen, J.A.M.Geelen, and M.P.J.G.Versleijen, “An improvement

deembedding technique for on-wafer high-frequency characterization,” in Proc.IEEE
Bipolar Circuit Technology Meeting, 1991, pp. 191-194.

[10] C.H.Chen and M.J.Deen, “A general noise and S-parameter deembedding

procedure foron-wafer high-frequency noise measurements of MOSFETs,” IEEE Trans.
Microw. Theory Techniques, vol.49, no.5, pp.1004-1005, May 2001.

78
[11] H.Cho and D.E.Burk, “A three-step method for the deembedding of high frequency
Sparameter measurements,” IEEE Trans. Electron Devices, vol.38, no.6, pp.1371-1384,
Jun 1991.

[12] Y.Cheng, M.J.Deen and C.H.Chen, “MOSFET Modeling for RF IC Design,” IEEE
Trans. Electron Devices, vol.52, no.7, July. 2005.

[13] A.F. Tong, K.S. Yeo, L. Jia, C.Q. Geng, J.-G. Ma and M.A. Do, “Simple and
accurate extraction methodology for RF MOSFET valid up to 20 GHz,” IEEE
Proc.-Circuits Devices Syst., Vol. 151, No.6,December 2004

[14] Jin, X., Ou, J.-J., Chen, C.-H., Liu, W., Deen, M.J., Gray, P.R., and Hu, C.: “An
effective gate resistance model for CMOS RF and noise modeling”. Technical Digest of
International Electron Devices Meeting, Dec 1998, pp. 961–964

[15] Cheng, Y., Chen, C.H., Enz, C., Matloubian, M., and Deen, M.J.: ‘MOSFET
modeling for RF circuit design’. Proc. 3rd IEEE Int. Caracas Conf. on Devices, Circuits
and Systems (ICCDCS 2000), Cancun, Mexico, Mar 2000, pp. D23/1–8

[16] Cheng, Y.H.: ‘MOSFET modeling for RF IC design’, Int. J. High

Speed Electron Syst., 2001, 11, (4), pp. 1007–1084

[17] Y.Cheng et al., “A Physical and scalabel BSIM3v3 I-V Model for Analog/Digital
Circuit Simulation,” IEEE Trans. Electron Devices, vol.44, pp.277-287, Feb, 1997.

[18] C.Hu et al, “BSIM3v3.2.2 MOSFET Model User’s Manual,” Univ. California
Berkeley, 1999.

[19] M.Kang, I.M.Kang, and H.Shin, “Extraction and Modeling of Gate Electrode
Resistance in RF MOSFETs,,” in IEEE Int.Conf.Integrated Circuit and Technology,
pp.207-210, 2005

[20] Y.Cheng, “MOSFET Modeling for RF IC Design,” Int. J. High Speed. Electron.
Syst., vol.11, no.4, 2001.

79
[21] T.Matsuoka, “VLSI summer school ～ CMOS RF circut session ～,” in
http://www6.eie.eng.osaka-u.ac.jp/reference/VLSI summer20050817.pdf , 2005.

[22] Noboru Ishihara and Toshihiko Shimizu, “A Simple Equation Model for RF

MOSFET,” Silicon Monolithic Integrated Circuits in RF Systems, 2006,Jan.2006

80
Acknowledgement
This thesis is written under the direction of Professor Hiroshi Iwai, Department of
Electronics and Applied Physics, Interdisciplinary of Graduate School of Science and

Engineering, Tokyo Institute of Technology. I would like to express my sincere and

deep gratitude to Professor Hiroshi Iwai for his through instructions, continuous

supports and encouragements.

I would like to thank to Professor Kazuya Masu, Associate Professor Kazuo Tsutsui,

Visiting Professor Yoshihiro Arimoto, Assistant Professor Kuniyuki Kakushima,

Assistant Professor Kenichi Okada, Assistant Professor Parhat Ahmet, Visiting

Professor Hitoshi Aoki, Visiting Professor Takeo Hattori, and Visiting Professor

Nobuyuki Sugii for their critical comments and valuable discussions.

This work was partially supported by Semiconductor Leading Edge Tehnologies, Inc.

(Selete). I would like to thank Mr.Yasuo Nara, Mr. Mitsuo Yasuhira , Mr. Fumio

Ohtsuka, Tsunetoshi Arikado, and Mr. Kunio Nakamura for their great supports in

researches as well as valuable discussions.

This work was partially supported by Matsushita Electric Industrial Co., Ltd.. I would

like to thank Mr. Hiroshi Shimomura, and Mr. Kenji Ohata for their great supports in

financials, researches, as well as valuable discussions.

Furthermore, I woul like to thank to former RF members, Mr. Satoshi Yoshizaki

(Analog Devices, Inc), and Mr..Chong Woei Yuan (Fuji Xerox Co., Ltd).

I would like to appreciate my colleagues Jin-Aun NG, Molina Reyes Joel, Atsushi

Kuriyama, Hendriansyah Sauddin, Yasuhiro Shiino, Takashi Shiozawa, JaeYeol Song,

Kiichi Tachi, Koji Nagahiro, Manabu Adachi, Yoshihisa Ohishi, Takamasa Kawanago,

81
Soshi Sato, Yasuhiro Morozumi, Koichi Okamoto, Issui Aiba, Yusuke Kuroki, Kentaro

Nakagawa, Akira Fukuyama, Satoshi Yoshizaki, Xiang Ruifei.

I would like to appreciate laboratory secretaries, Ms. Nobuko Iizuka, Ms. Yuki

Hashizume, Ms. Akiko Matsumoto, Ms. Takako Fukuyama , Ms. Mikoto Karakawa for

their supports.

Finally, I express my warm heartfelt to my father Toshiaki and my mother Yukiko,

elder sister Rie, and younger sister Akiko for their everlasting supports, encouragements

and understanding.

Masayuki Nakagawa

Yokohama, January 2007

82
83